Brain and Cognition PSYCH131 Lecture 6
30.10.18
Concept Formation and Categorisation
We need concepts to understand what is happening around us. Without perception, object
recognition and categorisation we would be unable to interact meaningfully with the world.
Categorising the things surrounding us, combined with our stored conceptual knowledge,
allows us to plan our actions
Object Recognition:
■ Inferotemporal neurons sometimes fire in response to a particular object.
BOOTH + ROLLS found some macaque neurons are view angle. This is
computationally difficult, the retinal image projected by the same object can change
dramatically with view angle
We can discriminate between categories but can also generalise to one group.
These two images show that we can both distinguish between individual exemplars (each
distinct letter A or separate shell) - DISCRIMINATION whilst we can also group perceptually
different exemplars into common categories - GENERALISATION (as all being the letter A or
as all being shells)
Object constancy: How do we consistently categorise objects to achieve invariance when
the same object can appear so different when we re-encounter it from a different view?
Occlusion – part of the object is hidden
Can self-occlude
Visual Object Constancy:
The visual system must recognise familiar categories of objects (identifying these as all
being staplers) whilst ignoring irrelevant variation in the input that we perceive (due to, for
example, depth rotations, plane rotations, and size, position and lighting changes) which
may make objects look different from one instance to the next
This is crucial as it lets us access the same semantic information (for example, that staplers
are used in offices) whatever view of the object we see
Concept - a mental representation of a class of things
(a type, eg "dog"; words refer to concepts) - the basis of our semantic knowledge
Category - the set of things that belongs to the concept class
(eg every actual dog that exists in the world belongs to the dog category)
Exemplar (or instance) - one of the set of things in a category
(eg my own pet dog, Smudge)
Attribute (or feature or property or characteristic) - something which can be true or false
of a particular thing (an exemplar) or of a class of things (category)
(eg the thing or class of things is red, or expensive, or angry, or alive)
, Typicality - either the distance in representational space of an exemplar to the category
prototype or the average distance of an exemplar to all other exemplars from the same
category
Similarity - the distance in representational space between either two exemplars or
two categories
Types of category
- natural - categories in the natural world
(eg tulip, volcano, giraffe, mushroom, milk, pebble)
- artefact - categories made by humans to serve a function or a purpose
(eg hammer: for hitting things, chair: for sitting on; bicycle: for transport)
- nominal or ad hoc - a category of things sharing only an arbitrary feature or characteristic
rather than having a meaningful set of common features
(eg the category of yellow foods, or of things you usually take on holiday, or of towns
with names ending in "y")
Theories of Categorisation:
1. Classical/ Defining features theory
2. Prototype
3. Exemplar
4. Explanation-based
1. Classical/ Defining Features
- Assume concepts are defined by a rule specifying the necessary features which
are jointly sufficient to decide if an item belongs to a category.
- If you are lacking a trait or something, you are not in the category. E.g. Bachelor
is unmarried adult male, if married, not, if woman, not, if child, not.
- Testing classical theories:
Concept acquisition (Bruner et al., 1956) – an
example cards given with feedback (+/- for
cards that belonged/didn't belong to the
category) until the participant correctly guessed
the rule that defined the concept
- Bruner et al. (1956):
- - role of memory - if only one card was shown
at a time this task was very hard. People were
better at working out the rules if they could see
all the cards at once (as here) or if they could
use pen and paper
- - role of strategy - people try different
strategies depending on their own preferences
and abilities as well as the time available and
whether they are constrained by their short-
term memory
- He found people can learn explicit rules that define concepts.
30.10.18
Concept Formation and Categorisation
We need concepts to understand what is happening around us. Without perception, object
recognition and categorisation we would be unable to interact meaningfully with the world.
Categorising the things surrounding us, combined with our stored conceptual knowledge,
allows us to plan our actions
Object Recognition:
■ Inferotemporal neurons sometimes fire in response to a particular object.
BOOTH + ROLLS found some macaque neurons are view angle. This is
computationally difficult, the retinal image projected by the same object can change
dramatically with view angle
We can discriminate between categories but can also generalise to one group.
These two images show that we can both distinguish between individual exemplars (each
distinct letter A or separate shell) - DISCRIMINATION whilst we can also group perceptually
different exemplars into common categories - GENERALISATION (as all being the letter A or
as all being shells)
Object constancy: How do we consistently categorise objects to achieve invariance when
the same object can appear so different when we re-encounter it from a different view?
Occlusion – part of the object is hidden
Can self-occlude
Visual Object Constancy:
The visual system must recognise familiar categories of objects (identifying these as all
being staplers) whilst ignoring irrelevant variation in the input that we perceive (due to, for
example, depth rotations, plane rotations, and size, position and lighting changes) which
may make objects look different from one instance to the next
This is crucial as it lets us access the same semantic information (for example, that staplers
are used in offices) whatever view of the object we see
Concept - a mental representation of a class of things
(a type, eg "dog"; words refer to concepts) - the basis of our semantic knowledge
Category - the set of things that belongs to the concept class
(eg every actual dog that exists in the world belongs to the dog category)
Exemplar (or instance) - one of the set of things in a category
(eg my own pet dog, Smudge)
Attribute (or feature or property or characteristic) - something which can be true or false
of a particular thing (an exemplar) or of a class of things (category)
(eg the thing or class of things is red, or expensive, or angry, or alive)
, Typicality - either the distance in representational space of an exemplar to the category
prototype or the average distance of an exemplar to all other exemplars from the same
category
Similarity - the distance in representational space between either two exemplars or
two categories
Types of category
- natural - categories in the natural world
(eg tulip, volcano, giraffe, mushroom, milk, pebble)
- artefact - categories made by humans to serve a function or a purpose
(eg hammer: for hitting things, chair: for sitting on; bicycle: for transport)
- nominal or ad hoc - a category of things sharing only an arbitrary feature or characteristic
rather than having a meaningful set of common features
(eg the category of yellow foods, or of things you usually take on holiday, or of towns
with names ending in "y")
Theories of Categorisation:
1. Classical/ Defining features theory
2. Prototype
3. Exemplar
4. Explanation-based
1. Classical/ Defining Features
- Assume concepts are defined by a rule specifying the necessary features which
are jointly sufficient to decide if an item belongs to a category.
- If you are lacking a trait or something, you are not in the category. E.g. Bachelor
is unmarried adult male, if married, not, if woman, not, if child, not.
- Testing classical theories:
Concept acquisition (Bruner et al., 1956) – an
example cards given with feedback (+/- for
cards that belonged/didn't belong to the
category) until the participant correctly guessed
the rule that defined the concept
- Bruner et al. (1956):
- - role of memory - if only one card was shown
at a time this task was very hard. People were
better at working out the rules if they could see
all the cards at once (as here) or if they could
use pen and paper
- - role of strategy - people try different
strategies depending on their own preferences
and abilities as well as the time available and
whether they are constrained by their short-
term memory
- He found people can learn explicit rules that define concepts.
